Journal Pre-proof

Introducing mathematical modeling to Estimate

Pavement Quality Index of Flexible Pavements
based on Genetic Algorithm and Artificial Neural
Networks

Shadi M. Hanandeh

PII: S2214-5095(22)00123-1
DOI: https://doi.org/10.1016/j.cscm.2022.e00991
Reference: CSCM991

To appear in: Case Studies in Construction Materials

Received date: 24 October 2021
Revised date: 27 February 2022
Accepted date: 28 February 2022
Please cite this article as: Shadi M. Hanandeh, Introducing mathematical
modeling to Estimate Pavement Quality Index of Flexible Pavements based on
Genetic Algorithm and Artificial Neural Networks, Case Studies in Construction
Materials, (2021) doi:https://doi.org/10.1016/j.cscm.2022.e00991
This is a PDF file of an article that has undergone enhancements after acceptance,
such as the addition of a cover page and metadata, and formatting for readability,
but it is not yet the definitive version of record. This version will undergo
additional copyediting, typesetting and review before it is published in its final
form, but we are providing this version to give early visibility of the article.
Please note that, during the production process, errors may be discovered which
could affect the content, and all legal disclaimers that apply to the journal pertain.
© 2021 Published by Elsevier.
Introducing mathematical modeling to Estimate Pavement Quality Index of
Flexible Pavements based on Genetic Algorithm and Artificial Neural
Networks

Shadi M. Hanandeh

Assistant Professor, Department of Civil Engineering, Al-Balqa Applied University, Email:

f
oo
[email protected]

pr
Abstract
e-
Pr
This study explored the outcomes of utilizing genetic algorithms and artificial neural
networks to assess the pavement quality index on the principal road by analyzing 500 flexible
pavement sections in Amman, Jordan. Pavement sections are selected in areas that are
al

exposed to many variables, such as traffic, pavement materials, and different climatic zones.
Pavement deterioration is determined by a number of factors, including cumulative equivalent
n

single axle load, pavement structure, and material properties. The study aims to develop a
ur

performance model of PQI based on using surface rating (SR), present serviceability rating
(PSR), and pavement age. Several techniques were used to propose the PQI model, such as
Jo

multiple linear regression, genetic algorithm, and artificial neural network. Multiple linear
regression showed that PSR and SR had a statistically significant influence only on PQI (P=
0.0001). However, age is less significant on PQI (p = 0.506). The genetic algorithm and the
artificial neural network techniques were applied to propose two PQI models with an R2
value for the training model of 0.98 and 0.94, respectively. The study results show means that
the genetic algorithm model performs better than the neural network.

Keywords: Genetic Algorithm, Artificial Neural Network, Pavement Management System,

Pavement Quality Index.
1. Introduction

The Pavement Management System is a collection of activities that include planning, design,
construction, maintenance, and evaluation of pavement. Pavement management systems
(PMS) employ available financial resources to select the optimum type of maintenance
depending on pavement conditions. A Maintenance Management System (MMS) is a type of
PMS that focuses on keeping the pavement in satisfactory condition during its life by
optimizing the available financial and technical resources. In addition, it is important to
predict pavement behavior during its lifecycle to provide rehabilitation before it reaches

f
oo
critical conditions. Therefore, performing MMS can decrease maintenance and rehabilitation
costs while also providing a safe transportation network.

pr
In Jordan, the amount of traffic on the primary roads has increased in the last 10 years, which
has caused dramatic deterioration on these roads. Therefore, better design, evaluation, and
e-
maintenance strategies are required. Amman Municipality is responsible for the construction
and maintenance of the road network in Amman. Since Amman Municipality has limited
Pr
funds from the Jordan government, effective maintenance and repair strategies are needed.
Pavement evaluation methods, i.e., Pavement Quality Index (PQI), are required in order to
al

develop these strategies and the PMS system. PQI is a composite of indexes that gives the
best establishment of pavement condition since it depends on the roughness of the surface and
n

pavement distress. Therefore, in order to perform a better M&R (Maintenance and

ur

Rehabilitation) treatment, a pavement performance model can be used to find the PQI under
different traffic loads over time. Jordanian roads are exposed to continuous deterioration
Jo

because of several factors influencing the pavement. The rate of deterioration on primary
roads increased because of high traffic loads accompanied by other factors such as aging,
quality of materials and construction, and environmental conditions. In addition, different
pavement performance models are being developed to obtain the most significant factors that
affect the performance and integrity of these roads. Two procedures of pavement condition
evaluation were being used in this study, pavement quality index (PQI) and Present
Serviceability Rating (PSR). Present Serviceability Rating (PSR) is a subjective technique
that represents the average choice made by five road users based on their riding comfort.
Also, pavement deterioration models were used to predict how the pavement would perform
in the future and how the pavement would look after different M & R activities. Pavement
performance models are the basis for evaluating current pavement conditions to select the best
approach for maintenance. While the pavement Condition Survey (PCS) is considered a
central concern in PMS, it also provides the required database to generate pavement
performance models. Corresponding to the importance of PCS, several procedures were
developed during the last century, such as paver system, serviceability rating, and roughness
elevation. Pavement distress detection and assessment have benefited from the development

f
of advanced technology [1]. For instance, an automatic system (CrackIT) integrated system

oo
was used for automated crack detection and characterization of flexible pavement surfaces
[2]. Another study was conducted based on fuzzy theory and the analytic hierarchy process, a

pr
decision-making approach for pavement prioritization using structural and functional
performance parameters [3]. Another study explained that modeling using artificial neural
e-
networks has more accuracy than classical multiple linear regression in pavement condition
assessment and management [4].
Pr
Skid resistance, structural adequacy, surface distress, and roughness are used to evaluate the
condition of the pavement (ride quality).The first two parts are seldom employed because of
al

their high cost, but the remaining elements are typically used by agencies. Also, road distress
n

may be assessed visually or by driving along the road, then analyzed utilizing distress
identification guidelines. Engineers with expertise use their eyes to determine pavement
ur

surface issues. For pavement condition surveys, road networks must be separated into sections
Jo

and sample units where the division method is used to get more precise data. The first
application of evaluating pavement condition occurred in the United States, and the first
model was developed in the mid-1970s by the Washington State Department of Highways [5–
7]. This model provides predictive performance and cost forecasting models based on a
database collected over a period of six to eight years in the state [8]. After the Washington
State model, several state departments of transportation started to generate their own PMS
based on their needs.

PMS can be used to evaluate pavement conditions and make maintenance recommendations.
There are several benefits of using PMS, which include saving a huge amount of money over
time, providing an effective and regular maintenance program, delivering a comprehensive
database including data related to inventory such as pavement condition, traffic, M&R, and
predicting the future requirements of the network pavement overtime [9]. Bosurgi and Trifirò
2005, developed an economic maintenance model for flexible pavement resurfacing using
ANN and genetic algorithm techniques. The findings demonstrated that policies based solely
on pavement friction increase can practice interference on pavement performance and provide
more reliable outcomes in reducing the number of accidents[10]. Elhadidy et al. 2021, utilized
a new model to link Pavement Condition Index (PCI) with International Roughness Index

f
(IRI) [11]. Fuentes et al. 2020, mentioned that destructive or non-destructive pavement

oo
structural assessment provides valuable information to road agencies about the structural
support, load-carrying capacity, remaining service life, pavement surface condition, and the

pr
selection of appropriate maintenance and rehabilitation (M & R) techniques for a variety of
road conditions. Most often, pavement layer moduli, which are back-calculated from
e-
deflection data, are utilized for both project and network level evaluations of the structural
state of the pavement. The use of "seed" moduli quantities during back calculation assessment
Pr
may also cause convergence to an optimum, which may result in subjectively incorrect final
moduli conclusions in certain cases. A pavement management model was used to predict IRI
al

values based on pavement management, which was assessed using condition ratings, and a
new equation between IRI and PCI was proposed [12]. Majidifard et al. 2020, developed a
n

pavement condition index using a deep machine learning approach. Based on the outputs of
ur

the hybrid model, two Pavement Surface Evaluation and Rating (PASER) prediction models
were provided by using Weight-based prediction and linear regression method. For all of the
Jo

data types, including training and testing data, performance is justified by measuring values of
R². Sensitivity analysis results showed that block cracking is the most practical parameter in
describing the differences in Pavement Surface Evaluation and Rating (PASER) as associated
with additional predictor variables. This may be because block cracking is a precise symbol of
high-level asphalt age and hardening[13]. Fuentes et al. 2021, used Pavement Condition Index
(PCI) and the International Roughness Index (IRI) to construct unique deterministic and
probabilistic methods that linked pavement serviceability to pavement indicators (IRI). In the
case of pavement condition, PCI is a numeric rating assigned to the pavement performance
based on the kind and intensity of distresses present on the surface, whereas IRI is an
indication of pavement roughness[14]. Also, Fuentes et al. 2021, assessed the user's
subjective experience of the degree of comfort on a road pavement and vertical accelerations
by analyzing deterministic and probabilistic models for pavement maintenance estimates,
which are applicable to urban roads with modest operating speeds. Current techniques for
evaluating road ride comfort are based on laboratory test findings, in part because laboratory
test circumstances are more likely to be rigorously controlled, resulting in more similar
results. Field experiments do not cause substantial vibrations on their own[14].

f
It was noticed that IRI, PSR, and PCI are values that indicate the pavement surface

oo
evaluation. To achieve a sophisticated and complete evaluation of the pavement surface, a
composite pavement performance is a rule necessity that needs to be implemented. Based on

pr
highway standards, (PQI) has been used for evaluating the pavement surface situation. The
PQI is an integrated index of PCI, Ride Quality Index, Skid Resistance Index and Rut Depth
e-
Index. PQI is used by many agenesis in the United States and Canada to evaluate pavement
condition because it represents the quality of road serviceability after maintenance activities
Pr
[15–17]. In the 1950s, the highway department in Minnesota developed a modified rating
system. In 1965, a new procedure, Condition Rating (CR), was developed. CR is an integrated
index that consists of two parts; the first part represents pavement surface condition, and the
al

other part is the ride quality of determined pavement. CR is calculated by averaging the PCR
n

and SR [15]. The purpose of measuring CR is to estimate the design life of roads for planning
ur

and programming the required needs. Thus, the design factors of flexible or rigid pavement
are appropriate and determine the priorities of maintenance needs. CR has been changed to
Jo

PQI in order to give a better indication of pavement condition than average condition.
Consequently, PQI models include parameters that represent pavement distress and pavement
roughness, which varies from state to state as shown in Table 1. Other states developed many
PQI models, including Ohio, Washington, D.C., Arizona, and Alberta, Canada), as shown in
Table 1.
 Table 1 Selected studies with proposed correlations

Number Equation Reference

1 ( ) SDI: Surface Distress [15]

Index

2 PQI = √( ) PSR: Present Serviceability Rating [16]

SR: Surface Rating

3 [17]
PQI ={

f
oo
( )

( )
PCR: Pavement Condition Rating

pr
4 [18]
( ) ( )

5
( ) (
e- )
[19]
Pr
6 [20]
al

7 [21]
( ) ( )
n

8 [22]
ur

{
( )( )
Jo

9 ( ) ( ) ( ) ( ) ( )( ) [23]

10 ( ) ( ) [24]

11 ( ) [25]

Bosurgi and Trifirò 2005, used artificial intelligence and Genetic Algorithm (GA) methods for
resurfacing disturbances on flexible pavements. This study found that the method describes an
effective procedure to achieve the best clarifications and decisions[10]. Lin et al. 2005,
investigated the relations between pavement distresses and IRI using an artificial neural
network methodology [26]. Several studies presented the ability to utilize ANN for pavement
evaluation [27–32]. An ANN was generated by Elhadidy et al. in 2021, to determine IRI from
PCI by using data gathered from the LTPP data. The generated ANN model produced an R2
of 0.86 which designates adequate prediction [11]. Domitrović et al. 2018, investigated
feasibility of utilizing ANN to estimate current pavement situations, and its ability to
determine maintenance strategies. Results indicated that ANN could be successfully used for
optimizing maintenance priorities [33]. Plati et al. 2016, used ANN to provide innovative
technique to predict pavement maintenance strategies [34]. Bosurgiet al. 2019, examined and
analyzed the ability of ANN to estimate the pavement performance index. The results

f
indicated that the number of hidden layers had significant effect on the model accuracy [35].

oo
Inkoom et al. 2016, used recursive partitioning and ANN in predicting pavement crack rate.
The results indicated that this technique improves the accuracy of prediction of pavement

pr
cracks condition ratings [36]. Hafez et al 2019, Used AAN with pattern recognition algorithm
for modeling the variability of maintenance activities. The results indicated that ANN based
e-
on pattern recognition algorithms are very beneficial permanent maintenance decision [37].
Pr
Current service rating (PSR), surface rating (SR), and Pavement Quality Index (PQI) are all
terms used in the construction industry. Each index represents a distinct element of pavement
condition and may be used to identify pavement divisions as well as anticipate future
al

maintenance and rehabilitation needs. Surface categorization (SR) Pavement difficulties are
n

obvious flaws in the pavement's surface. Cracks and permanent deformation are examples of
indicators that suggest a problem or degradation phenomenon in the pavement. The kind and
ur

degree of pavement deterioration can offer a detailed picture of future maintenance and/or
Jo

rehabilitation requirements. The SR is stated to the nearest tenth and varies from 0.0 to 4.0. A
higher SR indicates better pavement performance. A road that requires extensive restoration
or renovation will often have an SR of close to 2. The Pavement Quality Index is a composite
parameter that equals the square root of the product of PSR and SR, as a results SR gives an
overall assessment of the pavement's state.

2- Study Area

Amman's road network consists of more than 2500 kilometers of paved highways. These
highways are critical for the secure transportation of goods and people inside and beyond
Amman. Asphalt pavements cover more than 99 percent of the city's roadways [38]. Roads in
this region are subjected to a variety of unfavorable weather conditions, such as dense rain.
These causes induce premature pavement distress all year. As a result, roads in this location
undergo significant distress, such as load and environmental cracks, and permanent
deformation in pavement (rutting). This study investigated a total of 500 flexible pavement
sections with a width range of between 3 and 4 m, and about 40 m in length were evaluated.

2-1 DATA COLLECTION

Data about the roughness and cracks of the road surface, as well as rutting and cracks, were

f
investigated and quantified. The pavement condition data is utilized to evaluate system

oo
performance, help in project selection, and identify any future repair and/or rehabilitation
needs. Every year, pickup trucks travel the whole length of the roadway in both directions.

pr
2-2 Road subdivision features
Six distinct roads were chosen for this preliminary study, which is being presented in this
e-
article as shown in table 2. A total of 20 km of highway were examined, with more than half
Pr
of them being important arterial routes with heavy traffic all year. Present Serviceability
Rating (PSR), Surface Rating (SR), pavement age since last major maintenance, and (PQI)
data collected on specific pavement sections were examined for further analysis and data
al

interpretation. In order to calculate PQI, this study used the Minnesota procedure to find SR.
Each section was investigated and counted distress and the amount of severity. Then, the
n

amount of each type of distress was converted to a percentage by dividing the length of the
ur

measured distress by the length of the section 40m or area. Furthermore, multiply the
percentage of each distress multiplied by the appropriate weighting factor listed in the table to
Jo

get an individual weighted distress. Finally, using Fig 1, convert Total Weighted Distress
(TWD) to Surface Rating (SR).

Fig. 1. flexible Pavement Weighting Factors[39]

PSR estimation was performed for the 500 sections by dividing them into panels, and each
panel was driven by a passenger car over selected pavement sections at 80-90 km/hr. Each
panel member recorded his opinions on the scale of 0 to 5. The mean of the individual five
ratings was taken as the PSR of each selected pavement section. Drivers provide their
viewpoint on a range, from 0 to 5, wherever 5 indicates the best pavement situation and
performance, and 0 indicates extremely poor pavement performance. Table 2 is a summary of
the data that was gathered during this investigation, as well as a physical representation of the
overall road situation.

f
Table 2 SR, PSR and PQI data for selected sections

oo
Street Name SR Observation PSR Observation PQI Observation
4
Mecca Street 3.7 Very Good Very Good 3.8 Very Good

pr
Medina Street 3.2 Good 2.9 Fair 3.04 good

Wasfi Al-Tal Street 2.4

e-
Fair 3 good 2.68 Fair
Pr
Abdullah Ghosheh Street 3.58 Good 3.1 Fair 3.33 good

Queen Zein Al Sharaf Stree 2.7 good 3.2 Good 2.93 Good
al

Queen Alia International Airport 2.8 good 3.5 Good 3.13 Good
n

The principal purpose of the study is to predict a mathematical model to evaluate the
ur

pavement quality index (PQI) by using machine learning schemes. To achieve these
objectives, an integrated database that includes information about traffic, conditions, and
Jo

geometric characterization was developed. Geometric characterization includes road surface

conditions. In this study, two models were proposed by using genetic algorithms and artificial
neural networks. The input parameters were included as: pavement serviceability rating
(PSR), pavement age, surface rating (SR), and output variable PQI. A summary of the
statistical analysis data is presented in Table 3, and a sample of the selected data that was used
in this study is shown in Table 4.
Table 3 Statistics analysis for data used in models
Parameter No. of data Max Min Mean
SR 500 4.16 0.72 2.99

PSR 500 4.36 0.6.3 3.03

AGE 500 21 2 6.81

PQI 500 4.25 0.67 2.96

f
oo
Table 4 Sample data used in this study

Age SR PSR PQI

pr
3 3.68 3.33 3.50
3
13
3.61
2.76
e- 2.92
1.72
3.25
2.18
Pr
3 3.08 2.48 2.76
6 3.64 3.30 3.47
al

6 3.82 3.91 3.86

11 2.23 1.90 2.05
n

6 3.82 3.99 3.90

ur

9 2.78 2.99 2.88

6 2.41 1.64 1.99
Jo

6 3.42 3.07 3.24

4 3.82 3.85 3.83
4 3.83 3.94 3.88
4 3.83 3.86 3.83
8 3.23 3.18 3.20
9 3.31 2.48 2.86
10 3.45 3.78 3.60
3 3.61 3.53 3.56
3 3.56 3.28 3.41
3 3.53 3.58 3.55
8 2.69 1.92 2.70
8 3.74 3.30 3.51
8 3.13 2.89 3.00
8 3.83 3.78 3.80
8 2.75 2.51 2.62
3 3.76 3.56 3.66

f
oo
3 3.83 3.88 3.85
11 1.91 1.12 1.46

pr
4 3.72 3.08 3.39
5 3.13 2.68 2.90
2 3.83
e- 3.85 3.84
2 3.83 3.94 3.89
Pr
4 3.40 3.35 3.37
2 3.83 3.64 3.73
al

7 2.54 2.58 2.56

n

2 3.60 3.56 3.57

3 3.56 3.42 3.48
ur

7 2.37 1.82 2.07

Jo

6 1.80 3.60 2.55

3. Methodology

The data was analyzed using a multiple linear regression, genetic algorithm and an artificial
neural network. A genetic algorithm is a theoretical algorithm that relies on historical data,
uses the survival of the fittest notion. This strategy works well for many engineering fields
that have a wide range of properties[40],[41]. ANN is a high-performance model that
correlates system components based on numerical input, processing, output, and experience.
Pavement engineers employ the ANN model to address and forecast complex system
challenges. To keep the ANN viable, it must be paired with engineering skills. ANNs, like
human brains, learn via experience, examples, and practice[42]. Fig. 3 depicts the paper's
methodological flow chart.

f
oo
pr
Fig. 3. Methodological flow chart
3.1. Multiple linear regression (MLR)
e-
Pr
Multiple linear regression MLR is a technique that attempts to predict the relationship
between more than two explanatory (independent variables) and response (dependent)
variable. The interpretation of the coefficients in multiple linear regression equation is
al

estimated by the change of the dependent variable to a unit variety in the independent
variable, where the other independent variables are constant. Equation 1 shows the general
n

multiple linear regression model, where y is the dependent variable, βo is the intercept or
ur

constant, β1, β2, βn are coefficient of the variable X1 X2,…,Xn.

Jo

( )

To further establish the PQI behavior between SR, PSR, and age in the model, the model was
conducted several times with all the variables until the optimal model was achieved.
Regression showed that PSR and SR exerted significant influence only on PQI (p 0.0001).
However, age exerted less significant influence on PQI (p = 0.506). The multiple linear
regression (MLR) provides a good estimation with R^2 value of 98 as shown in equation 2.
Stepwise regression was performed on another model to avoid the less significant age in the
model and produced a new model with input variables of SR and PSR as shown in equation 3.
 - .1 2 . . - . 2 ge (2)

- .21 . 1 . 1 (3)

3.2. Application of Artificial Neural Networks

Artificial Neural Networks considered as a confidential method of modeling systems that

include several layers, input layer with bias, hidden layer with bias, and output layers. It is
therefore necessary to model advanced correlations between independent variables (inputs)
and dependent variables (outputs) to get a precise model from the collected data. Also, it's

f
oo
comprised of interconnected adaptation-easily process parts called "artificial neurons" or
"nodes" that are capable of generating art calculations for process and information illustration.
Artificial Neural Networks (ANN) illustrate the general relations of the arrangements on the

pr
side of the numeric coefficient that are tuned to support expertise, inputs, processes, and
e-
outputs. The design of the associate degree ANN model may be an assortment of nodes
distributed over the associate degree input data, hidden activation layer, associate degree, and
Pr
output data. When many trials and error rules were conducted, the simplest achievement was
captured at eight hidden layers, three hidden layers, and two hidden layers in three different
models. MATLAB was utilized to perform the ANN analysis. Fig. 4 displays the shape of the
al

neural model, as shown in the left layer to the right. The hidden layer's range is changed, and
n

with this changing by trial and error, a good model can be proposed.
ur
Jo

Fig. 4. Neural network arrangement

3.1.2. Estimate PQI by using ANN (Model 1)

As illustrated in Fig. 5, the creation of the neural model 1 is shown on the left side, with a set
of input parameters that include the three parameters discussed before. The number of
neurons was equaled to three by varying the number of hidden layers until an optimal model
was obtained. PQI is represented by one neuron on the right side. Fig. 6 depicts observations
of the anticipated robust PQI versus measured PQI. Furthermore, the ANN model delivers
precise destination value estimates. The R2 score of 0.93 and 0.91 for training and testing,
respectively, indicates that the model

f
oo
provides a reasonable approximation.

pr
e-
Pr
al

Fig. 5. ANN structure representing input parameters, output of hidden layers, and bias
for model 1.
n
ur
Jo

Fig. 6. Regression scheme comparing projected PQI against experimental PQI using model 1
for training and testing data.
ANN can predict PQI with the use of experimental data and produce the following equation
by including the equations for bias and weight average for the input data, hidden , and output
result:
 2. - . - .
-( .2 )
( 1⁄1 e 1 e- 1 e- 1 e- ) (4)

-2. . ge . 1 . 2

. 1.2 ge .2 .
. 1.1 ge .22 . 2

. 1.2 ge .2 .
. 1.1 ge .22 . 2

f
oo
3.1.3. Estimate PQI by using ANN (Model 2)

As illustrated in Fig. 7, the neural model 2 is being constructed, with many input parameters

pr
on the left side, which correspond to the two parameters previously discussed. The number of
neurons was equated to two by adjusting the number of hidden layers until the ideal model
e-
was produced. The PQI is represented by a single neuron on the right side of the brain.
Pr
n al
ur
Jo

Fig. 7. ANN structure representing input parameters, output of hidden layers, and bias for
model 2.
Fig. 8 shows observations of the anticipated robust PQI versus measured PQI. Furthermore,
the ANN model provides precise value of 0.95 and 0.94 for training and testing, respectively.
Fig. 8. Regression scheme comparing projected PQI against experimental PQI using model 2
for training and testing data.

f
Including the equations for bias and weight average for the input data, hidden and output

oo
result, ANN can predict PQI with the use of experimental data and produce the following
equation:

pr
. 1 . 1
-[1 .1 ]
1⁄1 e 1 e- 1 e-
e- (3)

. .2 .1
Pr
1. . .

3.3. Genetic Algorithm

al

The genetic algorithm method is a research style stirred up by biological process tools like
n

alternative, a different approach to function exploration, which will suitably meet the
ur

gathering of lab information. A genetic formula may be a particular improvement method that
produces robots for the purpose of solving problems by employing the assumption of
Jo

Darwinian generative ascertainment. Friedberg began the earliest steps within the field of
genetic formulas by employing a coaching formula to improve proposed model achievement.
A genetic formula produces a range of measures that represent the solution to any downside.
The organization of the genetic algorithm model is illustrated in Fig. 9.
 f
oo
Fig. 9. Genetic algorithm flow chart
Genetic software was invented by Cramer NL and further developed by Koza JR. Gene

pr
expression software (GeneXpro5.2) is employed to evaluate Pavement Quality Index (PQI).
During this study, organic phenomenon tree were accustomed to conducting investigations for
e-
gathered information. This programming package could be an effective machine from genetic
algorithm data wherever it operates to unravel several issues and build several models
Pr
mistreatment massive quantities of complex information and realize system in very specific
artificial languages like java and visual basic.
al

The origin of the practice of this network code relies on the number and quantity of
chromosomes, and it's comprised of gene numbers and expression trees, which specifically
n

the genetic info encoded within the body. (GeneXpro5.2) program is the most suitable to
ur

estimate Pavement Quality Index (PQI). Furthermore, by establishing a relationship between

the SR, PSR and age in model 1 and then between SR, PSR, in model 2. The program makes
Jo

many attempts to get the best solution once the creation has been selected, as if it were a
matter of altogether trails. The path was finished and optimized for the best model that
supported particular patterns like (R^2>0.99).

3.3.1. Estimate PQI by using GA (Model 1)

The construction of the GA model includes the PSR, SR and age. By changing the number of
number of chromosomes until the optimal model is obtained. Fig. 10 depicts observations of
the anticipated robust PQI versus measured PQI. Furthermore, the evolutionary algorithm
model delivers precise destination value estimates with R2 value of 0.97.
Fig. 10. Regression scheme comparing projected PQI against experimental PQI using model 1

f
oo
for training and testing data.

3.3.2. Estimate PQI by using GA (Model 2)

pr
The construction of the GA model includes the PSR, SR. By changing the number of number
e-
of chromosomes until the optimal model is obtained.Fig.11 depicts observations of the
anticipated robust PQI versus measured PQI. Furthermore, the evolutionary algorithm model
Pr
delivers precise destination value estimates with R2 value of 0.98.
n al
ur
Jo

Fig. 11. Regression plot of predicted PQI against experimental PQI for model 2 for training
and testing data.
3.3.3. Proposed equation by using GA

The GEP software has implemented design within the Illustration Tree kind wherever it's born-
again into mathematical kind exploitation. The rule method, as explained within the equations,
subsequently, interprets input variables and the additional constant values may be obtained from
the c++ coding of the GEP software. The proposed Pavement Quality Index (PQI) is the addition
of the three sub-trees (G1, 2, and 3). One of the benefits of GP methods is that the link between
model inputs and outputs may be expressed in the interpretation tree form, which is
automatically converted into numerical formulations. The formed GA paradigm is executed in
the manner of a tree explanation, as presented in Fig. 12 and 13 for the two models, respectively.
Three sub-trees were developed and converted into a scientific equation model as presented in
the equations below from equation
(4) 1( ub- -1) 2( ub- -2) ( ub- - ) (4)

f
oo
C8 = -0.09

pr
C9 = 1.37
C2 = 1.70
C4 = 9.48
C7 = 10.55
e-
Pr
C0 = 0.03
C2 = -1.91
d0= Age
al

d1= PSR
n

d2= SR
ur
Jo
 f
oo
pr
e-
Pr
n al

Fig. 12. Tree expression to estimate PQI from model 2

ur

The suggested GA method was used for all trial runs for regression procedures and the calculation of
Jo

the dependency of PQI. The resulting relationship between PQI and PSR, SR, and AGE characteristics
was found using the step-by-step approach outlined in equation 5.

ge .
( . 2 min ( , 1. )) (1. )
2 2 tan (min ((( . ) ( )),(( 1. 2 . ) ))) ( )
2 (1 . )

C8 = 2.07

C0 = -5.13
C5 = 0.91
C4 = 6.65
 C7 = -0.52
C2 = -3.91
d0= PSR
d1= SR

f
oo
pr
e-
Pr
n al
ur
Jo

Fig. 13.Tree expression to estimate PQI from model 1

The suggested GA method was used for all trial runs for both regression procedures and the
calculation of the dependency of PQI. The resulting relationship between PQI and PSR and
SR characteristics was found using the step-by-step approach outlined in equation 6.

( )
((( ) [ [ ]]) [ [ ( )
 ( ( ) ) ( )
] [( )] [ (

] (6)

4. Comparison performance of neural network and genetic algorithms models

There is no need for limitation or statistical interpretation to determine PQI throughout using
different pavement performance indices such as SR, SRA, and age. The genetic algorithm
produced higher R2 values, as well as lower χ2 and SSR values, as compared to the
mathematical model optimized via the neural network.

f
oo
2
( - )
(7)

pr
( )
quare est (χ2) ∑

(8)
e-
Pr
The findings demonstrate that the proposed genetic algorithm model's goodness of fit
indicators outperform the values produced by the neural network model as shown in table 5.
As a result, the suggested genetic algorithm model was chosen to represent greater R2values,
al

as well as lower χ2 and SSR values. The accuracy of the genetic algorithm model produced a
better fit and better outcomes, as seen in table 4. The GA model executes a regular estimation
n

and presents R2 for as 0.97 and for ANN R2 value for training as 0.94. This intimates that the
ur

genetic algorithms model presents more reliable estimates than ANN.

Jo

Table 4 Statistical results for Neural Network and Genetic Algorithm

Model 2 PQI
Model 1 PQI ANN Model 1 PQI GA Model 2 PQI GA
ANN
R2 0.93 0.94 0.97 0.98
SSR 39.58 36.99 16.16 9.13
Chi-Square
35.01 33.12 12.79 7.25
est (χ2)
5. Conclusion

This study recommended methods to estimate the pavement quality index on different roads
in Jordan. Two models of machine learning techniques, neural networks and genetic
algorithms, were employed to determine the pavement quality index with selected input
variables as follows: SR, PSR, and age. The artificial neural network was developed in Matlab
by employing different numbers of neurons in the hidden layer to obtain the optimum
achievement model with a high value of R2; the value of R2 for training and testing data
achieved by using Matlab programming software for the first model was 0.93 and 0.91,

f
oo
respectively. However, the results obtained from the genetic algorithm for model 1 were 0.97
for both the training and testing data. For the second model, the values of R2 for training and
testing data obtained by Matlab software were 0.95 and 0.94. However, the results obtained

pr
from the genetic algorithm for model 2 were 0.98 and 0.99, respectively. Because of the high
e-
value of the coefficient of determination for the proposed model, a gene expression algorithm
may be appropriate to resolve geotechnical issues and forecast the value of PQI with an
Pr
appropriate coefficient of determination (R2). One of the practical applications of the
developed models that can predict the PQI from simple input values is that these models
provide accurate results because they use advanced methods such as genetic algorithms and
al

neural networks. The merit of this study lies in the easy equations that are produced by this
n

software. The main limitation of this study is that the proposed models were produced from
the data within a specific range, as shown in table 3.
ur

References
Jo

[1] K. Gopalakrishnan, S.K. Khaitan, A. Choudhary, A. Agrawal, Deep Convolutional

Neural Networks with transfer learning for computer vision-based data-driven
pavement distress detection, Constr. Build. Mater. (2017).
https://doi.org/10.1016/j.conbuildmat.2017.09.110.

[2] H. Oliveira, P.L. Correia, Automatic road crack detection and characterization, IEEE
Trans. Intell. Transp. Syst. (2013). https://doi.org/10.1109/TITS.2012.2208630.

[3] V. Vyas, A.P. Singh, A. Srivastava, A decision making framework for condition
evaluation of airfield pavements using non-destructive testing, in: Airf. Highw.
 Pavements 2019 Innov. Sustain. Highw. Airf. Pavement Technol. - Sel. Pap. from Int.
Airf. Highw. Pavements Conf. 2019, 2019.
https://doi.org/10.1061/9780784482476.034.

[4] V. Vyas, A.P. Singh, A. Srivastava, Entropy-based fuzzy SWOT decision-making for
condition assessment of airfield pavements, Int. J. Pavement Eng. (2021).
https://doi.org/10.1080/10298436.2019.1671590.

[5] M.Y. Shahin, Pavement management for airports, roads, and parking lots: Second

f
edition, 2005. https://doi.org/10.1007/b101538.

oo
[6] R. Haas, W.R. Hudson, L.C. Falls, Pavement Asset Management, 2015.
https://doi.org/10.1002/9781119038849.

pr
[7] . uerra de Oliveira, . ibaut, . Dell’ cqua, irport pavement management
e-
systems: An open BIM approach, in: Lect. Notes Civ. Eng., 2020.
https://doi.org/10.1007/978-3-030-29779-4_44.
Pr
[8] M. Broten, Local Agency Pavement Management Application Guide, Northwest
Technol. Transf. Cent. (1996).
al

[9] A. Tavakoli, M.S. Lapin, J. Ludwig Figueroa, PMSC: Pavement management system
n

for small communities, J. Transp. Eng. (1992). https://doi.org/10.1061/(ASCE)0733-

947X(1992)118:2(270).
ur

[10] G. Bosurgi, F. Trifirò, A model based on artificial neural networks and genetic
Jo

algorithms for pavement maintenance management, Int. J. Pavement Eng. (2005).

https://doi.org/10.1080/10298430500195432.

[11] A.A. Elhadidy, S.M. El-Badawy, E.E. Elbeltagi, A simplified pavement condition
index regression model for pavement evaluation, Int. J. Pavement Eng. (2021).
https://doi.org/10.1080/10298436.2019.1633579.

[12] T.B. L Fuentes, K Taborda, X Hu, E Horak, A probabilistic approach to detect

structural problems in flexible pavement sections at network level assessment, Int. J.
Pavement Eng. (2020).
[13] H. Majidifard, Y. Adu-Gyamfi, W.G. Buttlar, Deep machine learning approach to
develop a new asphalt pavement condition index, Constr. Build. Mater. (2020).
https://doi.org/10.1016/j.conbuildmat.2020.118513.

[14] L. Fuentes, R. Camargo, G. Martínez-Arguelles, J.J. Komba, B. Naik, L.F. Walubita,

Pavement serviceability evaluation using whole body vibration techniques: a case study
for urban roads, Int. J. Pavement Eng. (2021).
https://doi.org/10.1080/10298436.2019.1672872.

f
[15] A. Transportation, Pavement design manual, Alberta, 1997.

oo
[16] S. Dai, E. Skok, T. Westover, J. Labuz, E. Lukanen, Pavement rehabilitation selection,
2008.

pr
[17] Flexible pavement condition survey handbook, 2009.

[18]
e-
DO ’s nhanced avement Management ystem, 2 .
Pr
[19] K.E. Stein-Hudson, R.K. Sloane, M.C. Jackson, A.J. Bloch, Customer-based quality in
transportation, 1995.

[20] N O ON NF U U etter Data Needed to ate the Nation’s

al

Highway Conditions, 1999.

n

[21] Federal Highway dministration, 2 2 tatus of the Nation’s Highways, ridges, and
ur

Transit: Conditions and Performance, Washington, D.C., 2003.

Jo

[22] P. VACURA, Pavement Smoothness Caltrans Challenges. California Department of

Transportation, in: Road Profile Users Gr. Meet., Stateline, Nevada., 2004.

[23] J. Nodes, moothness pecification rizona DO “ njoy the ide,” in: oad rofile
Users Gr. Meet., Stateline, Nevada, 2004.

[24] H.H. Titi, M. Rasoulian, Evaluation of smoothness of louisiana pavements based on

international roughness index and ride number, Jordan J. Civ. Eng. (2008).

[25] F. Reza, K. Boriboonsomsin, S. Bazlamit, Development of a Pavement Quality Index

for the State of Ohio, in: 85th Annu. Meet. Transp. Res. Board, Washington, D.C.,
2006.
[26] J. Lin, J.-T. Yau, L.-H. Hsiao, Correlation Analysis Between International Roughness
Index (IRI) and Pavement Distress by Neural Network, Transp. Res. Board 82th Annu.
Meet. (2003).

[27] H. Ceylan, M.B. Bayrak, K. Gopalakrishnan, Neural networks applications in

pavement engineering: A recent survey, Int. J. Pavement Res. Technol. (2014).
https://doi.org/10.6135/ijprt.org.tw/2014.

[28] A. Gupta, P. Kumar, R. Rastogi, Pavement deterioration and maintenance model for

f
low volume roads, Int. J. Pavement Res. Technol. (2011).

oo
https://doi.org/10.6135/ijprt.org.tw/2011.4(4).195.

[29] S. Ranjbar, F.M. Nejad, H. Zakeri, A.H. Gandomi, Computational intelligence for

pr
modeling of asphalt pavement surface distress, in: New Mater. Civ. Eng., 2020.
https://doi.org/10.1016/b978-0-12-818961-0.00003-x.
e-
[30] M.S. Yamany, T.U. Saeed, M. Volovski, A. Ahmed, Characterizing the Performance of
Pr
Interstate Flexible Pavements Using Artificial Neural Networks and Random
Parameters Regression, J. Infrastruct. Syst. (2020).
https://doi.org/10.1061/(asce)is.1943-555x.0000542.
al

[31] P. Marcelino, M. de Lurdes Antunes, E. Fortunato, M.C. Gomes, Machine learning

n

approach for pavement performance prediction, Int. J. Pavement Eng. (2021).

ur

https://doi.org/10.1080/10298436.2019.1609673.
Jo

[32] G. Bosurgi, O. Pellegrino, G. Sollazzo, Optimizing artificial neural networks for the
evaluation of asphalt pavement structural performance, Balt. J. Road Bridg. Eng.
(2019). https://doi.org/10.7250/bjrbe.2019-14.433.

[33] J. Domitrović, H. Dragovan, . ukavina, . Dimter, pplication of an artificial neural

network in pavement management system, Teh. Vjesn. (2018).
https://doi.org/10.17559/TV-20150608121810.

[34] C. Plati, P. Georgiou, V. Papavasiliou, Simulating pavement structural condition using

artificial neural networks, Struct. Infrastruct. Eng. (2016).
https://doi.org/10.1080/15732479.2015.1086384.
[35] G. Bosurgi, O. Pellegrino, G. Sollazzo, Road functional classification using pattern
recognition techniques, Balt. J. Road Bridg. Eng. (2019).
https://doi.org/10.7250/bjrbe.2019-14.448.

[36] S. Inkoom, J. Sobanjo, A. Barbu, X. Niu, Prediction of the crack condition of highway
pavements using machine learning models, Struct. Infrastruct. Eng. (2019).
https://doi.org/10.1080/15732479.2019.1581230.

[37] M. Hafez, K. Ksaibati, R.A. Atadero, Optimizing Expert-Based Decision-Making of

f
Pavement Maintenance using Artificial Neural Networks with Pattern-Recognition

oo
Algorithms, Transp. Res. Rec. (2019). https://doi.org/10.1177/0361198119851085.

[38] https://www.mpwh.gov.jo/Default/Ar, (n.d.). https://www.mpwh.gov.jo/Default/Ar.

pr
[39] Mn/DOT Pavement Distress Identification Manual, 2011.
e-
https://www.dot.state.mn.us/materials/manuals/pvmtmgmt/Distress_Manual.pdf.

[40] W. Al Bodour, S. Hanandeh, M. Hajij, Y. Murad, Development of Evaluation

Pr
Framework for the Unconfined Compressive Strength of Soils Based on the
Fundamental Soil Parameters Using Gene Expression Programming and Deep Learning
al

Methods, J. Mater. Civ. Eng. (2022). https://doi.org/10.1061/(asce)mt.1943-

5533.0004087.
n

[41] S. Hanandeh, A. Ardah, M. Abu-Farsakh, Using artificial neural network and genetics
ur

algorithm to estimate the resilient modulus for stabilized subgrade and propose new
Jo

empirical formula, Transp. Geotech. (2020).

https://doi.org/10.1016/j.trgeo.2020.100358.

[42] S. Hanandeh, S.F. Alabdullah, S. Aldahwi, A. Obaidat, H. Alqaseer, Development of a

constitutive model for evaluation of bearing capacity from CPT and theoretical analysis
using ann techniques, Int. J. GEOMATE. (2020).
https://doi.org/10.21660/2020.74.36965.
Declaration of interests

☒ The authors declare that they have no known competing financial interests or personal
relationships that could have appeared to influence the work reported in this paper.

☐The authors declare the following financial interests/personal relationships which may be
considered as potential competing interests:

f
There is no conflict of interest , and no any financial interests

oo
pr
e-
Pr
n al
ur
Jo